文献类型：期刊文献

中文题名：基于GPU平台和多源遥感的月度草畜平衡快速评价方法研究

英文题名：Study on Rapid Evaluation Method of Monthly Glass and Livestock Balance Based on GPU Platform and Multi-source Remote Sensing Data

作者：侯亚男[1] 张旭[1] 裴青生[2] 郭颖[1] 陈艳[1] 刘燕[1] 孙蕊[1] 杨铭伦[1]

第一作者：侯亚男

机构：[1]中国林业科学研究院资源信息研究所,北京100091;[2]青海大学畜牧兽医科学院,西宁810016

年份：2020

期号：3

起止页码：420-429

中文期刊名：科技促进发展

外文期刊名：Science & Technology for Development

收录：国家哲学社会科学学术期刊数据库;CSCD:【CSCD_E2019_2020】；

基金：2017年青海省畜牧兽医科学院重大科技专项(2017-NK-A4):海北州高寒草地生态畜牧业大数据管理平台与关键技术集成示范,负责人:张旭、郭颖

语种：中文

中文关键词：海北州;草畜平衡;合理载畜量;多源遥感;CatBoost;GPU

外文关键词：Haibei Tibetan Autonomous Prefecture;grass and livestock balance;carrying capacity;multi-source remote sensing data;CatBoost;GPU

分类号：TP79;S812.5

摘要：快速高频次评估区域草畜平衡状况,对于区域草地保护和经济发展起到重要作用。本文在GPU平台上结合多源遥感数据,构建了一种草畜平衡快速评估方法,并以海北州祁连试验区和海晏试验区2017—2019年6-9月的月度草畜平衡进行示范计算。结果表明:在牧草产量空间分布上,两试验区的分布范围随着时间的推移均呈现逐步减少的趋势;通过对月度单位面积产草量分析发现,两试验区单位面积产草量均在8月达到峰值,在9月开始下降;针对草畜平衡的分析,在2017—2019年间,祁连试验区每月均处于极度超载的情况,仅在每年8月份超载情况有所缓解;相比于祁连试验区,海晏试验区除在每年6月有少量超载情况外,7、8月基本能达到草畜平衡状态,在9月会出现载畜不足的现象。在草畜平衡快速估测效率方面,相比于随机森林和CatBoost-CPU模型,CatBoost-GPU模型可以更快地估测大区域范围内的产草量。本文可为当地有关部门及时调整放牧和补饲策略提供技术支撑。
To evaluate regional grass and livestock balance rapidly and frequently plays an important role in grassland protection and economic development. The paper built a method for dynamic monthly evaluation of the balance between grass and livestock by using multi-source satellite data based on GPU during June to September from 2017 to 2019 in the test sites of Qilian and Haiyan, Haibei Prefecture. The result shows that both test sites had a trend of gradually decrease for the spatial distribution of grass yield as time goes on. By the dynamic monthly analysis in both test sites, it was found that grass yield per unit area peaked in August and began to decline in September. In term of the balance between grass and livestock, the result showed that the situation was extremely overloaded every month and the overload status was relieved in August during 2017-2019 in Qilian test site. Compared with Qilian test site, the grass and livestock could be balanced in July and August every year, but it was overloaded in June and was insufficient livestock in September from 2017 to 2019 in Haiyan test site. For the evaluation result of the computation cost, the CatBoost-GPU model was faster in the large region grass yield estimation related to Random Forest and CatBoost-CPU. General speaking, the research could provide technological support for the local departments about the adjustment of the relationship between grazing and supplementary feeding timely.